Pass data from MetadataHook to BuildHook

[AzmainMahatab]

Description

What's the native way to pass data/objects created in MetadataHook to BuildHook?

Use case/motivation

If the data creation is expensive (e.g., version obtained via a network call), it doesn't make sense to re-create it in both steps, since the steps are sequential anyway.

Related issues

No response

Are you willing to submit a PR?

• Yes I am willing to submit a PR!

Code of Conduct

• I agree to follow the Python Software Foundation's Code of Conduct

[AzmainMahatab]

An even better suggestion is to join MetadataHook and BuildHook into a single class.
Why?
Currently, when you run hatch build --clean, it runs MetadataHook first, then BuildHook clean.
This sequence is not correct, as clean should happen at the very beginning, cleaning all previous files, providing a fresh start.
With this incorrect sequence, when you write any files to share with BuildHook, they get deleted with clean in the middle. clean should ensure the freshness of a session, not happen in the middle.

[cjames23]

An even better suggestion is to join MetadataHook and BuildHook into a single class.

These serve fundamentally different purposes in different phases of the build system. The scopes are different as well as their lifecycles. Metadata is run once during metadata resolution compared to BuildHooks that will need to run for sdist and wheel creation.

If the data creation is expensive (e.g., version obtained via a network call), it doesn't make sense to re-create it in both steps, since the steps are sequential anyway.

Can you please provide a concrete use case where you need to share data between the two?

[AzmainMahatab]

Metadata is run once during metadata resolution compared to BuildHooks that will need to run for sdist and wheel creation.

I think, even with a single combined class, achieving that is possible, but I understand that it might require a massive redesign and API change, which may not be feasible for the team.

Can you please provide a concrete use case where you need to share data between the two?

Suppose I make a Python wrapper that uses downloads and bundles a binary. When binary version "latest" is defined in toml, the whl version can only be set with MetadataHook, which requests the latest version tag. But I didn't find any native way to pass the version from the MetadataHook to the BuildHook. Requesting the latest version again can be 1. unnecessary cost, 2. theoretically incorrect, as the latest can very well update in between.

Finally, you didn't say anything about the order of cleaning in the sequence of hatch build --clean. Don't you think the sequence is wrong?

Currently, when you run hatch build --clean, it runs MetadataHook first, then BuildHook clean.
This sequence is not correct, as clean should happen at the very beginning, cleaning all previous files, providing a fresh start.
With this incorrect sequence, when you write any files to share with BuildHook, they get deleted with clean in the middle. clean should ensure the freshness of a session, not happen in the middle.

[cjames23]

Requesting the latest version again can be 1. unnecessary cost, 2. theoretically incorrect, as the latest can very well update in between.

This is mostly theoretical, the actual chance of this happening is so small that I am not sure that you will convince me that we should completely change the structure of hatchling for Metadata and Build hooks.

Finally, you didn't say anything about the order of cleaning in the sequence of hatch build --clean. Don't you think the sequence is wrong?

I did not say anything about this because first I wanted to understand your use case better before discussing too much further on solutions.